New York made companion chatbots name themselves and refer crises

The conversation

New York's General Business Law now answers a specific kind of conversation: one a user has with a companion chatbot late at night, in distress. Article 47 of the General Business Law, "Artificial Intelligence Companion Models," defines an AI companion as a system that retains information across sessions, asks unprompted emotion-based questions, and sustains an ongoing dialogue about matters personal to the user, including "intimate, romantic or platonic interactions or companionship." The New York State Senate's own law database shows this article's earliest recorded text dated 16 May 2025, with the version retrieved on 16 September 2026 revised as of 7 November 2025.

What the documents show

The statute imposes two concrete duties. Section 1702 requires an operator to give a clear and conspicuous notification at the start of an AI companion interaction, and again at least every three hours of continued use, stating that the user is not communicating with a human. Section 1701 makes it unlawful to operate an AI companion unless it contains a protocol for detecting and addressing expressions of suicidal ideation or self-harm, including a notification referring the user to crisis services such as the 988 hotline or a crisis text line. Section 1703 gives the state Attorney General authority to sue and to seek civil penalties of up to $15,000 per day for a violation of either duty, with collected funds directed to the state's suicide prevention fund.

The system boundary

The statute's own definition draws the boundary. Section 1700 excludes any system a business uses solely for customer service or to provide information about its commercial products, any system primarily designed and marketed for efficiency, research, or technical assistance, and any system used solely for internal business purposes. A task-oriented support bot is not automatically covered; the disclosure and crisis-referral duties attach specifically to systems built and marketed to sustain a personal, ongoing relationship.

Where it fails

The statute specifies the crisis-referral trigger only as "reasonable efforts for detecting" self-harm expressions, without describing a required detection method, leaving open how a regulator would judge whether an operator's effort was reasonable. It also leaves the boundary between companionship and customer service to be argued case by case as products blend both functions.

• Would our product's marketing and design place it inside or outside the statute's definition of an AI companion?
• Can we show, and log, that our disclosure appears at the start of a session and recurs within the three-hour window the law sets?
• What does our self-harm detection protocol actually do when it triggers, and does it name a real crisis-service referral rather than a generic disclaimer?

Where a product spans both categories, this reading of the statute's text is editorial; a firmer answer would come from the Attorney General's own enforcement record under Section 1703, which had not yet produced a public case as of retrieval.